"Also, compression algorithms normally produce binary data which would cause issues when used in string context (null characters) leading to, at a minimum, collisions"

But - if I get you right - for "seen-lookups" without ever tackling null characters, and eval{}-ed in - it should work?

"Also I don't think MD5 or SHA1 will buy you much savings giving the string lengths you're talking about."

Right. Some strings are even shorter than the 32bit MD5's. So the idea of not blowing up (MD5) but even compressing down (gzip) came up.